Hard drive failure calculator
Hard drives fail, even the best ones. What decides whether you lose your files is how many copies you keep, where, and how soon you'd notice one is gone.
How likely are you to lose your files?
Drive failure rates come from Backblaze's Drive Stats for Q2 2026: lifetime failure rates measured on over 350,000 drives in their data centres. "Best model" and "worst model" are the most and least reliable models in that table; "average" is all of their drives together. More on the numbers
What the fields mean
- Drive failures a year
- The chance that this drive dies in any given year. For a cloud backup, it's the chance of losing access to it instead. It's a blanket figure for every way of losing access to a cloud backup: a lost password or two-factor device, a lapsed payment, an account that gets locked or banned, the service shutting down, or the provider losing data on its side. It starts at 1% a year as a guess; set your own.
- Location
- Only matters if you include fire, flood and theft (03): copies in the same location are lost together. A cloud backup always counts as a location of its own.
- Checked
- For a drive you only connect now and then: how often you plug it in and look. If it has died in the drawer, you only find out the next time you check.
- Drives inside
- For a NAS: how many drives hold the files and how they're arranged. A mirror (RAID 1) or RAID 5 keeps the files when one drive dies, RAID 6 when two do, and striping (RAID 0) loses everything when any one drive dies. A dead drive is swapped and the array rebuilt in about the days to get it back. A fire, theft, power surge or mistake still takes the whole NAS at once: RAID is not a backup.
- Keeps old versions
- For copies that update by themselves: whether you can go back to yesterday's files. Without it, a deleted folder or ransomware is copied straight into the backup.
- Days to get it back
- Once you notice a copy is gone: how long until it's working again, meaning a new drive bought (or a new account set up) and your files copied back onto it. Until then you're one copy short, and if another copy fails in that time, you can lose everything.
The starting values come from US household figures, read generously: they include small fires and break-ins where nothing was taken. Change any of them to fit where you live; 0 switches a risk off.
A rough estimate from a simple model, not a prediction or a risk assessment. See the terms.
What would help most
Nothing on the list makes a real difference to this setup. It's already well protected against what it models.
Where the risk comes from
These odds count whole copies being lost. There's a quieter way to lose a file: it goes bad while every drive keeps working, and the damaged version gets copied into your backups. More on bit rot ↓
Where they come from
Drive failure rates are Backblaze's lifetime annualized failure rates from their Q2 2026 Drive Stats: the best model in their table fails at 0.5% a year, the fleet as a whole at 1.41%, the worst model at 5.88%. Those are drives in a data centre; yours may do better or worse.
The cloud figure, 1% a year, is a guess. It's a blanket figure for every way of losing access to a cloud backup: a lost password or two-factor device, a lapsed payment, an account that gets locked or banned, the service shutting down, or the provider losing data on its side. Nobody publishes how often that happens, so set your own.
The other risks start from US household figures: home fires from the NFPA, burglary from the Bureau of Justice Statistics, and lightning from Triple-I's count of home insurance claims, set against the Census count of owner-occupied homes. Each is read generously, and each is yours to change.
All sources, with the dates we checked them: see Sources ↓
Why a better drive matters less than a second copy: the most reliable hard drive? There's no perfect one.
If you want to read the drive tables yourself, there's a guide to reading Backblaze's drive stats on the FileGriffon blog.
What it assumes
- Each copy is OK, broken but not yet noticed, or being replaced. Your files are lost the first day no copy is OK.
- Failure rates stay the same over the years. Real drives fail more as they age, so long periods with old drives are worse than shown.
- Drives fail independently. Drives from the same batch can wear out together.
- A broken copy you check every month is noticed after about half a month on average, not on an exact day.
- In a NAS with several drives, the drives fail independently and a rebuild always succeeds. Real arrays do a little worse: drives from the same batch can fail close together, and a long rebuild can hit a second failure or a read error.
Bit rot: when a file goes bad quietly
A file can change while the drive holding it keeps working fine: a few bits flipped, a sector that no longer reads back right. No failure rate counts it, Backblaze's included, because no drive failed.
What makes it dangerous is the copies. Your next backup copies the damaged file over the good one, unless the backup keeps old versions from before the damage. More copies don't help on their own; what helps is noticing, by checking files against a record of what they should be, while a good copy still exists somewhere.
Checking files against a record can be done by hand with checksum tools, or automatically: FileGriffon, made by the people behind this site, does it on a schedule.
Some files can also carry recovery data that repairs damage without a backup at all. The parity bench shows how, one strip at a time.
Sources
Every number the calculator starts from, where it comes from, and when we last checked it. Anything not listed here is labelled as a guess on the page.
- Backblaze Drive Stats for Q2 2026, Backblaze, 2026. Drive failure presets: lifetime annualized failure rates of the best model (0.5%), all drives together (1.41%) and the worst model (5.88%), measured on 353,041 drives. Checked 10 November 2026.
- The Official Hard Drive Stats FAQ, Backblaze. What an annualized failure rate means, and how Backblaze chooses its drives.
- Home Structure Fires, National Fire Protection Association (NFPA). Fire: about 1 in 326 US households report a home fire each year (2012 to 2016 average), so 0.3% a year. Checked 10 November 2026.
- Criminal Victimization, 2023, Bureau of Justice Statistics (BJS). Burglary: 13.1 burglary or trespassing victimizations per 1,000 US households in 2023, so 1.3% a year. Checked 10 November 2026.
- Lightning-related homeowners claims, 2024, Insurance Information Institute (Triple-I), 2025. Lightning or surge: 55,537 US homeowners claims in 2024. Checked 10 November 2026.
- Owner-occupied housing units in the United States, US Census Bureau (Housing Vacancies and Homeownership), via FRED. About 86.4 million owner-occupied homes in 2024, used to turn the lightning claims into a yearly rate (about 0.06%). Checked 10 November 2026.
Not sourced, and labelled as guesses: losing access to a cloud backup, all causes together (starts at 1% a year), flood or water damage (starts at 0.05% a year), and mistakes or ransomware (starts at 1% a year).